Olfactory Perception Benchmark - Rate-All-That-Apply: leaderboard

Olfactory Perception task: select every odor descriptor that applies to a molecule from a 138-term lexicon. Per-question multilabel F1 in percent, higher is better.

Metric: Score (%) on rate-all-that-apply descriptor profiling (select every applicable descriptor from 138 for 100 molecules; per-question multilabel F1) of the 1,010 questions of the Olfactory Perception (OP) benchmark presented with compound-name prompts (the isomeric SMILES variant is reported only in figures), constrained option lists, provider APIs without web search or tools, each model at its stated reasoning effort or thinking budget; higher is better. Source: arxiv.org. Saturation forecast: Around 2031. 16 models tracked. Updated 7 Oct 2026.

Top models

#ModelSettingScoreOverall rank
1Claude Opus 4.5High42.2#160
2Claude Opus 4.6High40#113
3Grok 3 MiniHigh37#565
4GPT-5High36.4#178
5GPT-5 Pro36.1#163
6DeepSeek Reasoner (Thinking 16K) (Olfactory Perception checkpoint unspecified)36
7Grok 4.1 Fast35.5#427
8Claude Sonnet 4.534.9#261
9DeepSeek Reasoner (Thinking 8K) (Olfactory Perception checkpoint unspecified)34.7
10GPT-5.2 ProHigh34.6#137
11Gemini 2.5 ProThinking, 32K34#276
12DeepSeek Reasoner (Thinking 32K) (Olfactory Perception checkpoint unspecified)33.1
13O3High31.5#237
14O4 MiniHigh29#368
15Llama 3.3 70B Instruct26.8#1385

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=olfactory-perception-benchmark-rate-all-that-apply · How It Works · Data refreshed daily, snapshot 2026-11-21.